relationship between green space-related morphology and noise … · 2018. 5. 2. · margaritis...
TRANSCRIPT
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Relationship between green space-related morphology and noise pollution
Efstathios Margaritisa, Jian Kanga* aSchool of Architecture, University of Sheffield, Western Bank, Sheffield S10 2TN, United
Kingdom
** Corresponding author
Abstract Green spaces have been proved to have a positive effect on traffic noise pollution in the local scale; however their effects have not been explored on the urban level. This paper investigates the effects of green space-related parameters from a land cover viewpoint on traffic noise pollution in order to understand to what extent greener cities can also be quieter. A triple level analysis was conducted in the agglomeration, urban and kernel level including various case study cities across Europe. The green space parameters were calculated based on land cover data available in a European scale, while traffic noise data were extracted from online noise maps and configured in noise indices. In the first level 25 agglomerations were investigated, six of which were further analyzed in the urban and kernel levels. It was found that the effect of green spaces on traffic noise pollution varies according to the scale of analysis. In the agglomeration level, there was no significant difference in the cluster of the higher green space index and the percentage of people exposed in the lowest (55-59 dB(A)) or the highest noise band of more than 70 dB(A). In the urban level it was found that lower noise levels can possibly be achieved in cities with a higher extent of porosity and green space coverage. Finally, in the kernel level a Geographically Weighted Regression (GWR) analysis was conducted for the identification of correlations between noise and green. Strong correlations were identified between 60% and 79%, while a further cluster analysis combined with land cover data revealed that lower noise levels were detected in the cluster with higher green space coverage. At last, all cities were ranked according to the calculated noise index. Keywords : Noise pollution, Green spaces, Urban morphology, GIS, Cluster analysis 2016 Ecological Indicators Date Received: 20 May 2016 Date Accepted: 19 September 2016 Available online:6 October 2016
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
1. Introduction
The problem of exposure to traffic noise is rapidly increasing and is closely related with the
rapid urbanization process taking place around the world. Nowadays, 54 per cent of the
world’s population lives in urban areas, a proportion that is expected to rise to 66 per cent by
the year 2050 (United Nations, 2012). As a consequence of this process, noise annoyance
problems are caused, leading one out of five Europeans to be regularly exposed to sound
levels during the night that can trigger serious damage to health (WHO, 2009). This is the
reason why the European Community adopted measures for the noise reduction through the
Environmental Noise Directive (2002/49/EC), hereinafter called the “END”.
Other benchmarking reports on a European scale classified cities according to various
urban forms (Schwarz, 2010) or sustainability indices (The Economist, 2009). However, the
last report refers to transport variables, which cannot provide a direct assessment of the
noise pollution in these cities. From the viewpoint of soundscape, studies on a European
context are rare and there is the need to establish a common protocol for soundscape
exposure assessment (Lercher and Schulte-fortkamp, 2015). Lastly, in the European Green
Capital Award (European Comission, 2014), the quality of the acoustic environment was
taken into consideration using the exposure of people above or below certain noise bands
whenever these results were available.
On the other hand, green spaces comprise one of the inherent elements of urban form
apart from outdoor spaces, road and building infrastructure (Valente-Pereira, 2014). All
these factors can affect traffic noise distribution in various levels. Previous studies have
examined their effect either on the building level (Oliveira and Silva, 2010; Salomons and
Berghauser Pont, 2012; Silva et al., 2014) or in large neighbourhoods (Hao et al., 2015a;
Tang and Wang, 2007). At the city level, traffic noise has been measured either through the
use of landscape metrics (Oliveira and Silva, 2010; Mõisja et al., 2016; Weber et al., 2014)
or with the help of indicators related to road and building characteristics (Aguilera et al.,
2015; Hao et al., 2015b). Finally, on regional level, an attempt to approach noise issues by
emphasizing on the identification and designation of “quiet areas” according to land use
criteria was performed by Votsi et al., 2012.
The relationship between traffic noise and green spaces has been investigated in multiple
scales. The majority of these studies focuses on the small-scale, where the absorption or
scattering effects of branches and leaves are investigated (Attenborough, 2002; Aylor, 1972;
HOSANNA, 2014; Huddart, 1990; Van Renterghem et al., 2014). This kind of researches
cover a wide range from a single tree (Yang et al., 2011) to different plant types
(Horoshenkov et al., 2013) or various tree belts (Van Renterghem et al., 2012). Interesting
quantitative approaches on the park scale have also been developed by Pheasant et al.
(2010) with the Tranquillity Rating Prediction (TRAP) tool and by Brambilla and Gallo (2016)
with the QUIETE index. At the city scale, previous works have selectively emphasized the
quantitative assessment of parks concerning traffic noise reduction (Cohen et al., 2014;
González-Oreja et al., 2010). Other studies investigating also the users’ perception of the
acoustic quality in the parks have been performed by Brambilla et al., 2013; Brambilla &
Maffei, 2006; Weber, 2012.
However, there is little evidence on the effect of green spaces as a land use parameter on
traffic noise. The most frequent use is through land use regression (LUR) models (Goudreau
et al., 2014; Ragettli et al., 2016), or in a local scale through the TRAP tool by Pheasant et
al. (2010), which can be very useful in the absence of noise maps, but still of limited range
and dependent on on-site noise measurements.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Widely used indicators for green spaces usually refer to green space coverage (Fuller and
Gaston, 2009; Zhao et al., 2013) or green space per inhabitant (ISO 37120; WHO, 2010).
Others include also the proximity to green areas (Herzele and Wiedemann, 2003; Hillsdon et
al., 2006; Kabisch et al., 2016; Morar et al., 2014; Natural England, 2010; Ståhle, 2010) or
more complex indices referring to the balance between green and built up areas (De la
Barrera et al., 2016). Furthermore, there are shape-oriented indices, which can also
measure the distribution of green spaces (Margaritis and Kang, 2016; McGarical and Marks,
1994; Verani et al., 2015).
Consequently, the aim of this research is to provide, through the analysis of noise mapping
and land cover data, an evidence of whether greener cities can also be quieter. This
research question was investigated on three geographical levels (agglomeration, urban,
kernel) using a top-down perspective in order to investigate also the effect of the scale on
the results. The correspondent targets were: 1) the effect of forest, urban green and
agricultural areas on noise distribution in the agglomeration level, 2) the effect of green
space indicators on noise indices in the urban level and 3) the effect of green space
indicators on noise indices in the kernel level of the investigated cities.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
2. Methods
The methodology used investigates the relationship between green space and noise
indicators in three different levels starting from a general to a more focused scale. For
comparison purposes, the six cities namely: Antwerp, Helsinki, Brussels, Prague,
Amsterdam and Rotterdam mentioned in levels two and three also exist in level one. The
first part refers to the agglomeration level as defined in the END, while the second one refers
to the urban level, which is equal to or smaller than the administrative borders of the cities.
Finally, the third level refers to small kernel areas of 500x500 m each, covering the six cities.
It should also be made clear that the level of accuracy in these noise maps is acceptable for
this kind of strategic analysis, in spite of the differences in the production software or input
data, since all the results have to comply with the END requirements.
2.1 Agglomeration level
2.1.1 Case studies selection
Out of the available 216 agglomerations in the European Environment Information and
Observation Network database (EIONET, 2015), 25 were selected (12%) covering 11 out of
20 European countries. This was the maximum available sample size, since the selection
process was based on the availability of both noise mapping and land cover data for the
same agglomerations. The aim was to mostly cover medium-sized cities between 100,000
and 500,000 inhabitants, with bigger ones to serve as a means of comparison. The
population density of the sample as shown in Fig.1 has a broad range between 842 and
6,249 people / km2, while the population size of the agglomerations varies between 117,073
and 1,543,781 inhabitants. The agglomeration area ranges between 110 and 496 km2 as
presented in Table 1, while the green area coverage ranges between 35 and 405 km2.
Finally, the agglomeration borders were provided by the EIONET Agency as generalised
polygon shapefiles.
Fig. 1. Population density and average value (dotted line) in the agglomeration level. Source: EIONET
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Table 1
General characteristics of the 25 agglomerations sorted in a descending form for the population density field.
City Area (km
2)
Pop. Density (people/km
2)
Total Green (km
2)
Total Green (m
2/person)
Brussels 160 6,249 38 38
Valencia 130 6,249 94 115
Copenhagen 302 3,546 104 97
Helsinki 200 2,805 69 123
Sofia 492 2,760 116 85
Frankfurt 250 2,660 107 160
Tallinn 159 2,527 100 250
Lille 426 2,348 196 196
Prague 496 2,340 282 243
Hannover 238 2,333 96 174
Antwerp 205 2,062 35 83
Gratz 128 1,960 56 223
Linz 111 1,911 48 225
Varna 169 1,892 95 297
Montpellier 155 1,855 70 242
Amsterdam 152 1,752 46 173
Rotterdam 150 1,752 38 146
Alicante 200 1,674 136 406
Dresden 329 1,387 187 410
Grenoble 327 1,315 405 942
Ruse 127 1,240 80 508
Innsbruck 110 1,133 84 672
Burgas 219 1,050 204 884
Vitoria - Gazteiz 276 857 230 972
Bruges 139 842 71 602
2.1.2 Noise data and indicators
As Europe moves forward towards a common noise policy with harmonised noise
indicators; population exposure assessments can become a valuable tool of evaluating the
current and future noise conditions. The current data were sent to EIONET from the member
states through reports submitted in 2007 and updated until August 2013. Population
exposure is measured by the percentage of people affected per noise band using the Lden
index as mentioned in Table 2. This index was used in the current study in the absence of
original noise mapping data at the agglomeration level.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Table 2 Definition of variables related to noise (Source: EIONET) and green spaces (Source: Urban Atlas) in the agglomeration level.
Variables
Noise indices (% of people affected per noise band)
Green space indices
Lden(55-59)
Lden(60-64) Agricultural areas (%) Agricultural areas (m2/person)
Lden(65-69) Forest areas (%) Forest areas (m2/person)
Lden(>70) Urban green areas (%) Urban green areas (m2/person)
2.1.3 Green space data and indicators
Green spaces at this level are divided in three categories, namely: a) Agricultural areas, b)
Forest areas, and c) Urban green areas. These categories have already been defined in the
Urban Atlas land use dataset (European Environment Agency, 2010), where green space
data was downloaded from. Accordingly, the correspondent indices are expressed as the
coverage ratio per category and as the percentage per person (Table 2). From the noise
perspective, these areas represent the porous surfaces with higher sound absorption than
rigid surfaces.
The Urban Atlas land cover dataset is available for Large Urban Zones with more than
100,000 residents. The final data are provided in a scale of 1:10,000, while the original data
come from satellite images of 2.5m resolution, which is very precise for the analysis on a city
scale. However, it can also be used complementarily to other datasets such as CORINE
land cover, which makes the analysis easier and more comprehensive.
2.2 Urban level
2.2.1 Case studies
From the 25 agglomerations, six cities were selected as presented in Fig.2 in order to
perform a more detailed analysis on the urban level. The noise mapping area in these cases
is equal to or smaller than the agglomeration area, firstly because the main emphasis is in
the core city parts and secondly because agglomerations are abide by specific population
criteria (2002/49/EC).
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Fig. 2. Degree of differentiation between the agglomeration and the noise mapping areas in the six cities. Source: Brussels Institute for Environmental management, Czech Republic: Ministry of Health, Netherlands: Calculation
of road traffic noise, Helsinki City Council, Antwerp City Council, Topographic Basemap: ESRI)
There were two criteria for the selection process: a) the city should have an available
online noise map, b) the noise map should be continuous and cover the entire region, not
only the major roads. According to Table 3, the population size of the selected cities ranges
between 464,009 and 1,160,641 inhabitants. Apart from Prague and Brussels the rest of the
cities are in the upper population limit of mid-sized cities (Mpopulation=520,651) based on the
classification criteria by Bolton and Hildreth (2013). Additionally, population density in five
out of six cases range between 2,340 and 3,715 people (Mdenisty=3,425) per km2 (Table 3).
Table 3
General characteristics of the cities in terms of size and population density.
Agglomeration Name
Agglomeration area (km
2)
Noise mapping area (km
2)
Population (noise mapping area)
Density (people/km
2)
Brussels 160 162 999,899 6,172
Amsterdam 219 152 564,664 3,715
Rotterdam 326 149 464,009 3,114
Helsinki 186 200 570,578 2,853
Antwerp 205 205 483,353 2,358
Prague 496 496 1,160,641 2,340
2.2.2 Noise data and indicators
In every noise map it was necessary to calculate the percentage of pixels belonging to the
different noise bands. For similarity purposes and in order to have comparable results
among all the cities five noise bands were defined as presented in Fig.3.
All maps were imported in ArcGIS and converted to a raster file of 10-meter grid resolution
through a supervised classification. The same grid size has been used for the noise maps
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
produced by the Department of Environment, Food and Rural Affairs (DEFRA, 2007). An
identical process was followed for the green space data. Then with the help of “Zonal
Statistics” tool it was rendered feasible to have the exact number of noise and green space
pixels per band.
Fig. 3. Representation of the noise maps for the six cities using a common noise band scale.
In this level seven different noise indices - as presented in Table 4 - were formulated and
tested in order to check which one can better describe the extent of noise pollution in the
cities. Overall, three main approaches were adopted. In the first one the main idea was to
compare the number of pixels in the marginal bands of 55 dB(A) and over 70 dB(A) in each
city. This was sorted out with different combinations as described in Δnoise 1-3. Another
group of indicators (Δnoise 4, Δnoise 6) involved also the intermediate noise bands between
60 and 70 dB(A). Finally, the last index includes all the noise bands in a weighted sum. This
index attributes inverse weights from 1 to 5 by enhancing the lower noise bands and
diminishing the importance of the higher ones. The identification of the most suitable noise
index is based on the highest correlation between noise and green space indices.
Apart from the above indicators, other parameters were also tested for possible
correlations with green space variables and proved unsuccessful. More specifically, these
include the Building and Road Coverage, as well as five classes of the road network
hierarchy (Motorway, Residential, Primary, Secondary and Tertiary) defined as ratios of the
total road length.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Table 4
Noise and green variables tested.
Variables Definition / Notes Formula
Ratio per noise band
p55 % of noise pixels - 0 - 54.9 dB(A) p55 = 55(i) / sum (noise pixels)
p60 % of noise pixels - 55 - 59.9 dB(A) p60 = 60(i) / sum (noise pixels)
p65 % of noise pixels - 60 - 64.9 dB(A) p65 = 65(i) / sum (noise pixels)
p70 % of noise pixels - 65 - 69.9 dB(A) p70 = 70(i) / sum (noise pixels)
p70p % of noise pixels - 70 - 99 dB(A) p70p = 70p(i) / sum (noise pixels)
p(x)(i), x=55,..70p number of pixels per noise band
Noise indices
Δnoise 1 Index proportional to noise (↑↑) Δnoise1 = p70p/p55
Δnoise 2 Index proportional to noise (↑↑) Δnoise2 = p70p / p55 + 70p
Δnoise 3 Index inversely proportional to noise (↑↓) Δnoise3 = sum (p55-p70) / p70p
Δnoise 4 Index inversely proportional to noise (↑↓) Δnoise4 = p55 / [average (p60 - p70p]
Δnoise 5 Index proportional to noise (↑↑) Δnoise5 = p70p / [(average (p55-p70p)]
Δnoise 6 Index inversely proportional to noise (↑↓) Δnoise6 = [(p55 / p70p) * (1 / sum (p60-p70)]
Δnoise 7 Index inversely proportional to noise (↑↓) Δnoise7 = [5*p55+4*p60+3*p65+2*p70+p70p]
Green space indices
Green Space Ratio (Δgsr) % total green spaces Δgsr = Green space surface / Sum area
Extent of porosity (Δporous) % porous to non-porous surfaces Δporous = Δgsr / (BCOV + RCOV)
Forest ratio (Δtrees) % green space classified as "trees" Δtrees = Area of trees / Sum area
Free field ratio (Δfree field) % green space classified as "free field" Δfree field = Area of free field / Sum Area
BCOV Buildings ratio BCOV = Area of buildings / Sum area
RCOV Road Coverage ratio RCOV = Road area / Sum area
2.2.3 Green space data and indicators
There are various classification typologies for green spaces, which vary between eight and
nine categories depending on the classification criteria (Bell et al., 2007; Panduro and Veie,
2013).
In this research, a first set of indicators was established referring to the green space
coverage in the cities; firstly as a ratio compared to the whole area (Δgsr) and secondly as a
percentage between porous and rigid areas (Δporous) as presented in Table 4. Green space
data were extracted by Mapzen (Mapzen, 2016), which uses the latest Openstreetmap
dataset under an open license.
A second set of indicators was formulated from the sound propagation perspective, where
noise attenuation is higher with the presence of trees and lower with grass or any other low
vegetation (ISO 9613-B). Subsequently, the indicators established were named as “Δtrees”
and “Δfree field”. The first one refers to areas with a predominant presence of trees such as
forests, nature reserves and orchards, while the second one involves lower vegetation with
grass, scrubs, allotments or parks. Finally, Openstreetmap was selected as a more
favourable dataset compared to Urban Atlas, since in the second case data were not
available for all cities.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
2.3 Kernel level
In the last level of analysis a more focused approach was followed using the previous six
cities so as to test the same correlations between noise and green in a larger scale.
Specifically for this analysis a Geographically Weighted Regression (GWR) approach was
applied, which can better describe the geographical variations between the variables instead
of assuming that a single linear model can be fitted to the entire study area (Bristol
University, 2009) The parameters used in the GWR tool in ArcGIS (ESRI, 2016a) include a
fixed kernel type combined with the AICc bandwidth method, which can identify the optimal
adaptive number of neighbours for each case study area. The produced output refers to a
multipart area composed of various 500x500 meter-pixel blocks.
In order to bring the noise and green space data in an applicable format for the GWR,
some steps had to be followed in advance. First of all, by using the Block Statistics tool in
ArcGIS (ESRI, 2016b), green space data were aggregated using the rectangular
neighbourhood option. The aggregation field distinguishes forest areas from free field areas
(grass) by applying a weight of “2” in the first case and a weight of “1” in the second case.
For the noise data the same tool was used calculating the average values of the cells. The
final output files ranged between 1 to 5,000 for the green areas and 55 to 80 dB(A) for noise
levels. However, the final resolution in both datasets was adjusted to 500x500m as depicted
in Fig.4 so as to keep a balance between precision and calculation time.
Fig 4. Example of the applied kernel (500x500m) in Prague (a) and Helsinki (b).
In the final step, a cluster analysis was applied in order to identify the character of each
cluster in terms of land cover characteristics. The Grouping Analysis tool (ESRI, 2016c) was
used for this purpose with no spatial constraints, which allows features to be grouped only
based on their spatial proximity. Practically this process works in the same way as the k-
means partitioning method. At last, the produced clusters were intersected with CORINE
land cover data (European Environment Agency, 2006) of the finest resolution (100m x
100m) in order to get a more precise idea concerning the spatial distribution of the clusters.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
3. Results
3.1 Effect of green spaces on noise at the agglomeration level.
The question investigated at this level is whether there is an effect of different green space
categories such as forests, urban green and agricultural areas on noise. For this reason a
first cluster analysis was performed in order to divide the agglomerations in groups of “high”
and “low” green space areas per person. The particular cluster analysis can make the
identification of correlations between noise and green easier, since direct linear relationships
between the two variables were not found.
For the identification of possible clusters within the agglomerations a hierarchical analysis
with the three green space categories - where population is also involved - was applied
using the Ward’s method in SPSS. The analysis of coefficients and the “elbow rule” showed
that the optimal number of clusters is two. According to this result, the 25 agglomerations
were classified in two groups of high and low green, as depicted in Fig.5a.
Fig. 5. Analysis in the 25 agglomerations: (a) Levels of green space per person in the two clusters, (b)
percentage of people within the 55-59 dB(A) noise band, (c) percentage of people over 70 dB(A) according to the hierarchical analysis.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
The first cluster contains six agglomerations (Alicante, Bruges, Burgas, Innsbruck, Ruse,
Vitoria-Gazteiz) with high percentage of agricultural and forest areas, while the urban green
is low. On the contrary, the second cluster with 19 agglomerations is more balanced among
the three categories with a slightly higher percentage of urban green, but lower average
green space area per person in the other two categories compared to the first cluster.
The first independent samples T-test was applied solely for the green space categories. It
was found that there was a significant difference in the mean values of agricultural
(t(23)=6.7, p=.002) and urban green (t(23)=-4.6, p=.002) between the two groups of cities.
On the contrary, there was no difference in the mean values of the forest areas (t(23)=-2.5,
p=.80). The second T-test was then applied in order to test the hypothesis that the
percentage of people exposed to the lowest noise band (55-59) and the cluster with the
higher percentage of green (cluster 1) are positively correlated. The same process was
followed in order to check whether there is a negative correlation between the percentage of
people exposed to the highest noise band (>70) and cluster 1.
Results from both tests proved that the variances between the two clusters were different
from each other (t(23)55-59=1.21, p=.23 and t(23)>70=-1, p=.32.), however these differences
were not statistically significant as it can also be seen in the box-plots of Figs.5b,5c. In spite
of this fact and taking into account the scale of analysis, there is a tendency, showing that
more people are inclined to live in cluster 1 (Fig.5b). Similarly in Fig. 5c it can be seen that
the majority of people living in areas of more than 70 dB(A) belongs within cluster 2, where
all green space indices are lower.
In an attempt to identify similarities in the characteristics of the agglomerations within each
cluster it was shown that cities in the first group have a population density lower than the
average. From a land use perspective, according to the Urban Atlas classes, these
agglomerations are mostly covered by “discontinuous low density urban fabric” mixed with
industrial activities around the core urban area. Moreover, these places are characterised by
a clear segregation between urban and green classes, with a low percentage of mixture.
On the contrary, the second cluster involves agglomerations with 43% higher population
density on average than cluster 1. A higher coverage in the “continuous” and “discontinuous”
dense urban fabric was also observed, which is expected due to the population density
increase. Finally, there is a higher percentage of mixture between green and urban classes
in contrast with the segregated landscape of cluster 1.
3.2. Effect of green space indicators on noise in the urban level
3.2.1 Trends between noise and green
Initially, a graphic representation of the variations between noise and green (Δtrees+Δfield)
was produced so as to identify possible common trends among the six cities. According to
Fig.6 three different trends in terms of noise and green can be recognised.
The first one refers to cities like Helsinki and Prague where there is a parallel decreasing
tendency both in the number of noise pixels and green for each noise band. Moreover, both
cities represent cases with a very high proportion of quiet areas, which belong in the 55
dB(A) band and the highest proportion of green in the same frequency. A variation of this
trend can also be identified in the cities of Amsterdam and Rotterdam, where the decreasing
tendency in noise and green starts from 60 dB(A) instead of 55 dB(A).
The second trend refers to cities like Brussels, where noise has a more normal distribution
among noise bands compared to the first trend. Moreover, green and noise follow opposite
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
tendencies in the middle noise bands demonstrating that the relationship between these two
indices is not always proportional.
Finally, the last trend refers exclusively to the city of Antwerp, since it presents a pattern,
which is opposed to the expected one as observed in the first trend. Cities in this category
have relatively high percentage of green spaces and high noise levels with small variations
in all bands. It is also interesting that in Antwerp noise and green present an increase also in
the highest frequency. One of the possible reasons for this profile is the fact that Antwerp is
also a Trans - European Transport Networks corridor with many highways and constant
traffic.
Fig. 6. Comparison of the percentage related to noise pixels and pixels related to Green Space Coverage (Δgsr)
for the six cities.
3.2.2 Correlations between noise and green space indicators
The effect of green space indicators on noise indices was investigated in the urban level by
using all the related variables (Table 4). A Pearson product-moment correlation coefficient
was computed to assess the relationship between the seven noise indices and the four
green space dependent variables. Results proved that there was a positive correlation for
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
two of them. In particular Δnoise4 was positively correlated with Δporous (r=.76, n=6,
p=.045) and Δgsr (r=.82, n=6, p=.023). Similarly Δnoise6 had a positive correlation with
Δporous (r=.79, n=6, p=.035) and Δgsr (r=.85, n=6, p=.016). The scatterplot presented in
Fig.7 summarizes these results. As Fig.7a shows lower noise levels - expressed with high
values of Δnoise4 (R2=.72) and Δnoise6 (R2=.80) - can be achieved with higher levels of
porous surfaces. Similar results can be achieved with an increase in the green space
coverage (Δgsr) as shown in Fig.7b reaching a high coefficient of determination (R2 >.90).
Fig. 7. Coefficient of determination (R2) between Δnoise4, Δnoise6 and (a) Δporous, (b) Δgsr
A simple linear regression model was then calculated to predict noise levels (Δnoise6)
based on Δporous and Δgsr. The formulated regression equation provided statistically
significant results (F(2,4)=25.1, p<.05) with an R2 of .92. The variable of Δporous had the
highest contribution in the model (R2=.62) and Δgsr contributes with an additional value of
R2=.30. Practically this means that the balance of porous surfaces in a city can possibly
contribute to the reduction of traffic noise through proper land use planning.
3.2.3 Ranking of cities based on the selected noise index
The selection of Δnoise6 as the most suitable noise index at the urban level provides the
opportunity to rank the case study cities from the “quietest” to the “noisiest”. The ranking
process among the cities as presented in Fig.8 showed that the less noise-polluted city at
this level is Prague, with Helsinki, Brussels, Amsterdam, Rotterdam and Antwerp to follow.
The results reveal that the sequence of cities according to the noise index is not always the
same with the order of cities based on the porosity index or the green space coverage.
Practically this means that quieter cities can potentially be greener, however this does not
always work vice versa. For example, Amsterdam appears quieter than Brussels; however
Brussels has a higher ratio of green space coverage (Fig.8).
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Fig. 8. Ranking of cities from the quietest to the noisiest and interaction with Δporous and Δgsr.
3.3 Effect of green space indicators on noise in the kernel level
At this level correlations between green and noise were tested for each city via a GWR
approach by applying a moving search window in kernels of 500x500m. The sample of
14,932 observations (kernels) was big enough to facilitate this process. Then results were
grouped into clusters in order to identify patterns between the green and noise variables. In
the final stage, the groups were intersected with land cover data for a more comprehensive
identification of the cluster characteristics.
At first, the corresponding results of the GWR presented in Fig. 9 gave significant
correlations between noise and green with an R2 range between .60 and .79. Such high
correlations indicate that the relationship between the two variables varies locally and is
more meaningful when analysed using a moving window approach with a fixed kernel. Prior
attempts to interpret the same relationships using an Ordinary Least Squares (OLS) linear
regression model provided insignificant results. Finally, as regards the cities, the highest
correlation was calculated for Rotterdam (R2=.79), while the lowest for Brussels (R2=.60).
Areas that present no results within the borders of each city represent kernels with no
intersection between noise and green space data.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Fig. 9. The effect of green on noise according to the results of the GWR model and the associated coefficient of
determination (R2) for each city.
3.3.1 Ranking of cities
In order to test the consistency of the noise index (Δnoise6), which was selected for the
analysis on the urban level, a similar approach was followed also for the kernel level. The
index was recalculated for each area of 500x 500m and the final results were averaged for
the entire cities. Results shown in Fig.10 present similarities and differences compared to
the corresponding ones for the urban level (Fig.8). Specifically, three cities, namely Brussels,
Rotterdam and Antwerp retained their ranking positions (3rd, 5th, 6th). On the contrary Prague
was moved from the first position to the fourth, while small changes were evident for
Helsinki, which was moved from the second to the first position. Finally, Amsterdam was
ranked second instead of the fourth position in the urban level. Overall, it seems that the
transition from one scale to the other had an impact on the noise assessment of the cities,
although robust results in half of the case studies prove that the index has the potential to be
consistent. Other parameters that were expected to have an effect on the final ranking
comparison include the transformation of noise levels from discrete to continuous values and
the selected size of the kernel (500x500). In all cases, these results can only provide a
general initial insight for each city, which can further be elaborated during the planning
process.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Fig. 10. Ranking of the cities from the quietest to the noisiest according to Δnoise6 calculated in the kernel level.
3.3.2 Cluster analysis in the kernel level
A cluster analysis was applied after the GWR results were obtained. This process can lead
to a better understanding of the kernel areas according to the correlations between noise
and green space indices. The optimal number of groups as presented in Fig.11a was equal
to 3 according to the results from the total “within sum of squares” plot with the number of
clusters. The graph in Fig 11b describes the balance of the two variables among each
cluster. What can be concluded is that cluster 1 is typical of high green space coverage and
low noise levels, while opposite characteristics are present for cluster 3. Lastly, cluster 2
presents a balanced amount of green and noise in lower proportions compared to the other
two clusters.
Fig. 11. (a) Number of clusters and variance explained (within groups sum of squares). The optimal number of
clusters is determined according to the “elbow rule”. (b) Cluster variations based on the balance between green and noise using standardised values.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
The grouping analysis as shown in Fig.12 presents the spatial distribution of the three
clusters in the case study cities. Areas representing “group 1” are typical of high green space
coverage and low noise levels. Such areas are more representative in Prague (46%),
Brussels (17%), Antwerp (16%) and Helsinki (15%), with fewer samples in the other cities.
Areas of “group 2” represent kernels with low green space coverage and also low noise
levels. This kind of places can be found in the majority of the territory in Helsinki (68%),
Amsterdam (60%), Rotterdam (59%) and Brussels (58%). Finally, areas of “group 3” with
high noise levels and low green space coverage were evident in all cities, however higher
proportions were identified in Antwerp (51%), Rotterdam (35%), Amsterdam (31%) and
Prague (30%).
Fig. 12. Cluster analysis based on the k-means algorithm performed with the Grouping Analysis tool in ArcGIS. Basemap Source: ESRI.
More comprehensive conclusions can be drawn when combining the results of the cluster
analysis with CORINE land cover data. The analysis as presented in Fig.13 revealed that
over 30% of the agricultural and forest areas belong to cluster 1 as well as a small amount of
the total urban areas (5%). Very low percentages were present in this group as regards
industry, infrastructure, the rest of the vegetation and the water bodies.
Cluster 2, which has low levels of noise and green was found to include the highest
percentage of total urban areas (21%) and very low proportions in all the other classes. The
relationship between noise and green was not so evident in this group or at least results
were poorly correlated even with a GWR approach. The highest amount of green spaces in
this group was found in the forest class (14%).
Lastly, cluster 3 has the highest percentage in industry and infrastructure and also the
lowest in forest areas. Urban areas constitute the class with the highest proportion in this
group (10%) as in the other two clusters and agricultural areas depicted a higher percentage
than cluster 2.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
In general, cluster 3 appeared to have a reduced amount of green spaces compared to
cluster 1. In particular, there was a reduction of 17% in forest and agricultural areas and 1%
in the rest of the vegetation. Overall it was shown that at least in the marginal clusters (1,3)
noise and green had an inversely proportional relationship.
Fig. 13. CORINE land cover classes in the six cities distributed over the three clusters.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
4. Conclusions
The purpose of this study was to investigate whether greener cities around Europe can
also be quieter and less noise polluted. For this reason an analysis was conducted
investigating possible correlations between green space-related indicators and traffic noise
indices. The analysis was applied in three levels (agglomeration, urban, kernel) from a
broader to a smaller scale. In the first level, 25 European agglomerations were selected,
while six of them were further investigated on the urban and kernel level using noise data
from online noise maps. In the kernel level, each one of the six cases previously mentioned
was divided in grid areas (kernels) of 500x500m and a GWR analysis was conducted. Apart
from the identified correlations, the kernels were further grouped in clusters and associated
with land cover characteristics. Conclusions can be summarized as follows:
In the agglomeration level cities were divided in two clusters based on three green space
categories and particularly the green area per person. The two groups of cities were found to
be significantly different in agriculture and urban green ratio, with cluster 1 to present the
highest proportions. However, there was not a direct correlation between green space
indices and the population exposed in low (55-59 dB(A)) or high (over 70 dB(A)) noise
bands. As a result, the hypothesis that the percentage of people, exposed in the 55-59 noise
band, would be higher in the cluster with the higher green space index was not confirmed.
The same happened with the hypothesis that the percentage of people exposed in more
than 70 dB(A) would be higher in the cluster with the lower green space index. However, in
both cases there were tendencies towards the validation of both hypotheses, since the
variances between the two clusters were different from each other in the T-tests.
Concerning the land use attributes in the two clusters, it was found that cluster 1 was
related to urban and industrial areas with low population density and high segregation
between the green and urban classes. On the contrary cluster 2 was associated with high
urban land cover and high population density, but lower segregation between green and
urban areas.
In the urban level it was proved that quieter cities can potentially be greener, however this
does not always work vice versa. On the top of that the analysis showed that lower noise
levels can possibly be achieved in cities with a higher extent of porosity and green space
coverage. Between the two variables, the extent of porosity was proved to have a higher
contribution (R2=.62) in the prediction of noise levels than the extent of green space
coverage (R2=.30). As regards the detected trends, three different kinds were found. The
first one refers to cities, which present a parallel decreasing tendency both in noise levels
and green starting from the lower and moving to the higher noise bands. The second trend
refers to cities where noise had a more normal distribution among noise bands compared to
the first trend. The last trend involves cities, which have a relatively high percentage of green
spaces and also high noise levels with small variations in all noise bands.
In the kernel level, significant correlations were identified with the GWR approach and the
coefficient of determination (R2) to range between 60% and 79%. The clusters, which were
formed by the data of all cities, showed that it was possible to classify kernels in three main
groups. The first group was typical of high green space coverage and low noise levels. The
second one presented a balance between noise and green, while the third one was typical of
high noise levels and low green space coverage. While ranking the cities based on the same
noise index (Δnoise6) recalculated for the kernel level, three of them namely Brussels,
Rotterdam and Antwerp retained their ranking positions and the rest mainly presented small
variations.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
The cluster analysis at this level gave a number of three optimal clusters with the first one
to present low noise levels and high green space ratio. The second one was more balanced
between the two attributes, while the third one was exactly opposite to the first.
A further comparison of the three groups with the land cover data showed that noise levels
were minimized in the group that had the highest percentage of forest and agricultural areas
in combination with the minimum coverage of infrastructure, such as road or rail network,
ports and airports. This cluster accounted for 23% of the total area in the six cities. On the
contrary, the third cluster with the highest noise levels was combined with the maximum
coverage of infrastructure and industrial land encompassing an area of 34%. Finally, the
second cluster - with a total coverage of 43% - was typical of an urban environment with the
highest proportion of urban land cover and low fluctuations in the other classes including
industry, infrastructure and greenery.
Overall, the transition from one level to the other showed that the relationship between
noise and green can vary. However, some core relationships especially in the urban and
kernel level remained unchanged. In particular, the negative correlations in the urban level
suggest that planners should emphasize more on the ratio between green space and built-
up surfaces, since it seems to be more meaningful as an indicator compared to the green
space coverage itself. The modifiable area unit problem (MAUP), while moving from the
urban to the kernel level was minimised thanks to the small kernel size (500x500m) and the
application of a moving window with a fixed kernel in the GWR. An Ordinary Least Square
(OLS) regression was also applied, however the correlations were really poor proving that
the relationship between noise and green cannot be represented by a single global model at
that level.
A further research, in particular on the clusters’ urban part in particular can reveal more in
depth correlations between noise and green space features for the core parts of the cities.
Specifically, the integration of noise mapping data in land use regression models can be
more effective for noise pollution prediction during the urban and sub-urban areas early
planning stage.
Acknowledgments
This research received funding from the People Programme (Marie Curie Actions) of the
European Union's Seventh Framework Programme FP7/2007-2013 under REA grant
agreement n° 290110, SONORUS "Urban Sound Planner". The authors are also thankful
to Ms Nuria Blames and the GIS department of EIONET and to the City Councils of
Helsinki and Prague for the noise and land use data allowance.
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
References
Aguilera, I., Foraster, M., Basagaña, X., Corradi, E., Deltell, A., Morelli, X., Phuleria, H.C., Ragettli, M.S., Rivera, M., Thomasson, A., Slama, R., Künzli, N., 2015. Application of land use regression modelling to assess the spatial distribution of road traffic noise in three European cities. J. Exposure. Sci. Environ. Epidemiol. 25, 97-105. doi:10.1038/jes.2014.61
Attenborough, K., 2002. Sound propagation close to the ground. Annu. Rev. Fluid Mech. 34, 51-82. doi:10.1146/annurev.fluid.34.081701.143541.
Aylor, 1972. Noise reduction by vegetation and ground. J. Acoust. Soc. Am. 51, 197-205.
Bell, S., Montarzino, A., Travlou, P., 2007. Mapping research priorities for green and public urban space in the UK. Urban For. Urban Green. 6, 103-115. doi:10.1016/j.ufug.2007.03.005.
Bolton, T., Hildreth, P., 2013. Mid-sized cities: Their Role in England’s Economy. Centre for Cities, London.
Brambilla, G., Gallo, V., 2016. QUIETE: a scheme for a new index of the environmental quality of green areas. Noise Mapp. 3, 49-58. doi:10.1515/noise-2016-0004.
Brambilla, G., Gallo, V., Asdrubali, F., D’Alessandro, F., 2013. The perceived quality of soundscape in three urban parks in Rome. J. Acoust. Soc. Am. 134, 832-9. doi:10.1121/1.4807811.
Brambilla, G., Maffei, L., 2006. Responses to noise in urban parks and in rural quiet areas. Acta Acust. united Ac. 92, 881-886.
Bristol University, 2009. Introduction to Geographically Weighted Regression. http://www.bristol.ac.uk/media-library/sites/cmpo/migrated/documents/gwr.pdf (accessed 09.09.16).
Cohen, P., Potchter, O., Schnell, I., 2014. A methodological approach to the environmental quantitative assessment of urban parks. Appl. Geogr. 48, 87-101. doi:10.1016/j.apgeog.2014.01.006.
Department of Environemt, Food and Rural Affairs (DEFRA), 2007. Noise Mapping England: Maps and Charts. http://services.defra.gov.uk/wps/portal/noise (accessed 08.06.15).
De la Barrera, F., Reyes-Paecke, S., Banzhaf, E., 2016. Indicators for green spaces in contrasting urban settings. Ecol. Indic. 62, 212-219. doi:10.1016/j.ecolind.2015.10.027.
ESRI, 2016a. Interpreting GWR results. http://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-statistics-toolbox/interpreting-gwr-results.htm (accessed 08.08.16).
ESRI, 2016b. Block Statisitcs. http://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-analyst-toolbox/block-statistics.htm (accessed 08.08.16).
ESRI, 2016c. Grouping Analysis. URL http://pro.arcgis.com/en/pro-app/tool-reference/spatial-statistics/grouping-analysis.htm (accessed 08.08.16).
European Commission, 2014. European Green Capital Award 2015 - Urban environment good practice and benchmarking report.
Environmental Noise Directive (END) 2002/49 of the European Parliament and the Council of 25 June 2002 relating to the assessment and management of environmental noise, 2002. Official Journal of the European Communities L189, 12-25.
European Environment Agency, 2006. Corine Land Cover raster data. http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster-2 (accessed 29.03.16).
European Environment Agency, 2010. Urban Atlas. http://www.eea.europa.eu/data-and-maps/data/urban-atlas (accessed 02.04.16).
European Environment Information and Observation Network (EIONET), 2015. Noise database http://forum.eionet.europa.eu/etc-sia-consortium/library/noise_database/index_html (accessed 11.03.16).
Fuller, R.A., Gaston, K.J., 2009. The scaling of green space coverage in European cities. Biol. Lett. 5, 352-355. doi:10.1098/rsbl.2009.0010.
González-Oreja, J.A., Bonache-Regidor, C., De La Fuente-Díaz-Ordaz, A.A., 2010. Far from the noisy world ?
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
Modelling the relationships between park size , tree cover and noise levels in urban green spaces of the city of Puebla , Mexico. Interciencia 35, 486-492.
Goudreau, S., Plante, C., Fournier, M., Brand, A., Roche, Y., Smargiassi, A., 2014. Estimation of spatial variations in urban noise levels with a land use regression model. Environ. Pollut. 3, 48-58. doi:10.5539/ep.v3n4p48.
Hao, Y., Kang, J., Krijnders, D., Wörtche, H., 2015a. On the relationship between traffic noise resistance and urban morphology in low-density residential areas. Acta Acust. united Ac. 101, 510-519. doi:10.3813/AAA.918848.
Hao, Y., Kang, J., Krijnders, J.D., 2015b. Integrated effects of urban morphology on birdsong loudness and visibility of green areas. Landsc. Urban Plan. 137, 149-162. doi:10.1016/j.landurbplan.2015.01.006.
Herzele, A. Van, Wiedemann, T., 2003. A monitoring tool for the provision of accessible and attractive urban green spaces. Landsc. Urban Plan. 63, 109-126. doi.org/10.1016/S0169-2046(02)00192-5.
Hillsdon, M., Panter, J., Foster, C., Jones, A., 2006. The relationship between access and quality of urban green space with population physical activity. Public Health 1127-1132. doi:10.1016/j.puhe.2006.10.007.
Horoshenkov, K. V, Khan, A., Benkreira, H., 2013. Acoustic properties of low growing plants. J. Acoust. Soc. Am. 133, 2554-65. doi:10.1121/1.4798671.
Huddart, L., 1990. The use of vegetation for traffic noise screening, Publication of: Transport and Road Research Laboratory. Crowthorne, Berkshire, England.
ISO 37120: Sustainable development of communities - Indicators for city services and quality of life, 2014. http://www.iso.org/iso/catalogue_detail?csnumber=62436 (accessed 12.03.16).
ISO 9613-B: Acoustics-Attenuation of sound during propagation outdoors - Part 2: General method of calculation, International Standard Organization, Geneva, Switzerland, 1996. http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=20649 (accessed 12.03.16).
Kabisch, N., Strohbach, M., Haase, D., Kronenberg, J., 2016. Urban green space availability in European cities. Ecol. Indic. 70, 586-596. doi:10.1016/j.ecolind.2016.02.029.
Lercher, P., Schulte-fortkamp, B., 2015. Soundscape of European Cities and Landscapes - Harmonising., in: Final Conference of Soundscape-COST-TD0804-Project. Merano. doi:10.13140/2.1.3030.1127.
Mapzen, 2016. Open source map platform. https://mapzen.com/ (accessed 12.01.16).
Margaritis, E., Kang, J., 2016. Relationship between urban green spaces and other features of urban morphology with traffic noise distribution. Urban For. Urban Green. 15, 174-185. doi:10.1016/j.ufug.2015.12.009.
McGarical, K., Marks, B.J., 1994. FRAGSTATS: Spatial pattern analysis program for quantifying landscape structure. Oregon State University, Corvallis.
Mõisja, K., Uuemaa, E., Oja, T., 2016. Integrating small-scale landscape elements into land use / cover : The impact on landscape metrics ’ values. Ecol. Indic. 67, 714-722. doi:10.1016/j.ecolind.2016.03.033.
Morar, T., Radoslav, R., Spiridon, L.C., Pǎcurar, L., 2014. Assessing pedestrian accessibility to green space using GIS. Transylvanian Rev. Adm. Sci. 42, 116-139.
Natural England, 2010. Nature Nearby, Accessible Natural Greenspace Guidance. http://publications.naturalengland.org.uk/publication/40004?category=47004 (accessed 05.10.15).
Oliveira, M. F., & Silva, L. T. (2010). How urban noise can be influenced by the urban form. Proceedings in Advances in Biology, Bioengineering and Environment 2010.Athens,Greece, pp.31-36.
Panduro, T.E., Veie, K.L., 2013. Classification and valuation of urban green spaces - A hedonic house price valuation. Landsc. Urban Plan. 120, 119-128. doi:10.1016/j.landurbplan.2013.08.009.
Pheasant, R.J., Horoshenkov, K. V., Watts, G.R., 2010. Tranquillity rating prediction tool (TRAPT). Acoust. Bull. 35, 18-24.
Ragettli, M.S., Goudreau, S., Plante, C., Fournier, M., Hatzopoulou, M., Perron, S., Smargiassi, A., 2016. Statistical modeling of the spatial variability of environmental noise levels in Montreal, Canada, using noise
Margaritis Efstathios and Jian Kang: Ecological Indicators doi: 10.1016/j.ecolind.2016.09.032
Ecological Indicators, Volume 72, 2017, pages 921-933
measurements and land use characteristics. J. Expo. Sci. Environ. Epidemiol. 1-9. doi:10.1038/jes.2015.82.
Hosanna.,2014. WP3: Trees, shrubs and bushes. http://www.greener-cities.eu/ (accessed 15.1.14).
Salomons, E.M., Berghauser Pont, M., 2012. Urban traffic noise and the relation to urban density, form, and traffic elasticity. Landsc. Urban Plan. 108, 2-16. doi:10.1016/j.landurbplan.2012.06.017.
Schwarz, N., 2010. Urban form revisited-Selecting indicators for characterising European cities. Landsc. Urban Plan. 96, 29-47. doi:10.1016/j.landurbplan.2010.01.007.
Silva, L.T., Oliveira, M., Silva, J.F., 2014. Urban form indicators as proxy on the noise exposure of buildings. Appl. Acoust. 76, 366-376. doi:10.1016/j.apacoust.2013.07.027.
Ståhle, A., 2010. More green space in a denser city: Critical relations between user experience and urban form. Urban Des. Int. 15, 47-67. doi:10.1057/udi.2009.27.
Tang, U.W., Wang, Z.S., 2007. Influences of urban forms on traffic-induced noise and air pollution: Results from a modelling system. Environ. Model. Softw. 22, 1750-1764. doi:10.1016/j.envsoft.2007.02.003.
The Economist, 2009. European Green City Index, Siemens AG. https://www.siemens.com/entry/cc/features/greencityindex_international/all/en/pdf/gci_report_summary.pdf (accessed 05.11.15).
United Nations, 2012. World Urbanization Prospects: The 2011 Revision http://www.un.org/en/development/desa/population/publications/pdf/urbanization/WUP2011_Report.pdf (accessed 05.02.16).
Valente-Pereira, L., 2014. Urban form definition in urban planning. Revolução eBook
Van Renterghem, T., Attenborough, K., Maennel, M., Defrance, J., Horoshenkov, K., Kang, J., Bashir, I., Taherzadeh, S., Altreuther, B., Khan, A., Smyrnova, Y., Yang, H.-S., 2014. Measured light vehicle noise reduction by hedges. Appl. Acoust. 78, 19-27. doi:10.1016/j.apacoust.2013.10.011.
Van Renterghem, T., Botteldooren, D., Verheyen, K., 2012. Road traffic noise shielding by vegetation belts of limited depth. J. Sound Vib. 331, 2404–2425. doi:10.1016/j.jsv.2012.01.006.
Verani, E., Pozoukidou, G., Sdoukopoulos, A., 2015. The effect of urban density, green spaces and mobility patterns in cities’ environmental quality: An empirical study of the metropolitan area of Thessaloniki. Spatium 8-17. doi:10.2298/SPAT1533008V.
Votsi, N.E.P., Drakou, E.G., Mazaris, A.D., Kallimanis, A.S., Pantis, J.D., 2012. Distance-based assessment of open country quiet areas in Greece. Landsc. Urban Plan. 104, 279-288. doi:10.1016/j.landurbplan.2011.11.004.
Weber, M., 2012. The soundscape of nature areas : assessment and review of research approaches, in: Proceedings of the Acoustics 2012. Nantes, France, pp. 2139-2144.
Weber, N., Haase, D., Franck, U., 2014. Traffic-induced noise levels in residential urban structures using landscape metrics as indicators. Ecol. Indic. 45, 611-621. doi:10.1016/j.ecolind.2014.05.004.
World Health Organization (WHO), 2009. Night noise guidelines for Europe.
World Health Organization (WHO), 2010. Urban planning, environment and health.
Yang, H., Kang, J., Cheal, C., 2011. Sound dispersion and reverberation by a single tree, in: Internoise. pp. 1-10.
Zhao, J., Chen, S., Jiang, B., Ren, Y., Wang, H., Vause, J., Yu, H., 2013. Temporal trend of green space coverage in China and its relationship with urbanization over the last two decades. Sci. Total Environ. 442, 455-465. doi:10.1016/j.scitotenv.2012.10.014.